A number with missing labels<\/h2>\n\n\n\n
The four-to-one figure sounds precise, but it is not yet a usable technical measurement. The transcript does not identify the exact workload, calculation format, model configuration, or whether Liang was discussing raw computing power, training throughput, inference speed, or an entire rack-scale system.<\/p>\n\n\n\n
Also Read: Nvidia sells one chip for every four Huawei chips, and DeepSeek’s CEO just explained why that gap won’t close<\/a><\/h4>\n<\/div>\n<\/div><\/div>\n<\/div>\n\n\n\nThe supplied report also attributes a larger estimate to Liang, saying a model with 800 billion active parameters could require 50,000 Nvidia GB300 chips or 200,000 Huawei 950 chips. Yet Huawei\u2019s own roadmap<\/a> says the full Atlas 950 SuperPoD, built around Ascend 950DT processors, is scheduled to become available in the fourth quarter of 2026. <\/p>\n\n\n\nThat makes the comparison partly forward-looking, not a public apples-to-apples test that independent researchers can reproduce today. <\/p>\n\n\n\n![\"High-performance](\"https:\/\/okdiario.com\/techy\/en\/wp-content\/uploads\/2026\/08\/nvidia-huawei-deepseek-ai-chips-gap.jpg\" "\\"\\"")
DeepSeek founder Liang Wenfeng highlights hardware constraints, noting a significant performance gap between Nvidia and Huawei chips.<\/figcaption><\/figure>\n\n\n\nThe power math is different<\/h2>\n\n\n\n
Chip count does not translate directly into electricity use. Nvidia says one liquid-cooled GB300 NVL72 rack<\/a> contains 72 Blackwell Ultra graphics processing units, or GPUs, and can require up to 142 kilowatts. <\/p>\n\n\n\nIf 50,000 GPUs were arranged in those full racks and operated at maximum draw, the result would be roughly 695 racks and 99 megawatts of IT equipment capacity before broader facility overhead is added.<\/p>\n\n\n\n
That calculation is only an illustration, not a verified DeepSeek installation. Huawei says a full Atlas 950 SuperPoD will connect 8,192 neural processing units, or NPUs, across 160 cabinets and occupy about 10,800 square feet, but its official announcement does not state a directly comparable total power figure. <\/p>\n\n\n\n
Four accelerators could use less, similar, or more electricity than one GB300 once utilization, memory, interconnects, cooling, and software efficiency enter the picture.<\/p>\n\n\n\n
More chips also mean more than a larger electric bill. They can require additional circuit boards, optical links, power equipment, cooling hardware<\/a>, replacement parts, and physical space. The environmental cost may not rise in a neat four-to-one line, but it rarely stops at the processor package.<\/p>\n\n\n\nAI\u2019s electricity curve is steep<\/h2>\n\n\n\n
The International Energy Agency estimates that data centers consumed around 415 terawatt-hours of electricity worldwide in 2024. Its base case projects that total to reach about 945 terawatt-hours by 2030, while electricity used by accelerated servers grows around 30% each year. AI is not the only cause, but it is the most important new driver. <\/p>\n\n\n\n
\n\n\nAlso Read: The US just placed 17 states under emergency order to stop blackouts as extreme heat overloads the grid<\/a><\/h4>\n<\/div>\n<\/div><\/div>\n<\/div>\n\n\n\nChina and the United States are expected to account for nearly 80% of global data center electricity-demand growth through 2030. That concentration matters because massive computing campuses connect to local grids, where generation and transmission projects move much more slowly than chip product cycles. <\/p>\n\n\n\n
A more efficient hardware and software path could cut data center demand by more than 15% in the IEA\u2019s high-efficiency scenario<\/a>. <\/p>\n\n\n\nScarcity shaped DeepSeek<\/h2>\n\n\n\n
DeepSeek\u2019s reputation was built on doing more with constrained hardware. Its official DeepSeek-V3 technical report describes a mixture-of-experts model with 671 billion total parameters and 37 billion activated for each token, trained with 2.788 million H800 GPU-hours. <\/p>\n\n\n\n
The company estimated the direct rental cost at about $5.58 million, while explicitly excluding earlier research and experimental work. <\/p>\n\n\n\n
That record helps explain why the leaked comments matter. DeepSeek\u2019s efficiency was not simply a green initiative or a clever marketing line. It was an engineering response to limited access to advanced chips, and the alleged transcript suggests Liang still sees raw computing power as the main constraint. <\/p>\n\n\n\n
There is a catch. Lower energy use per training run or generated token does not guarantee lower total consumption if cheaper AI leads to many more models, users, and queries. Efficiency can slow the growth curve, but it can also make demand easier to expand.<\/p>\n\n\n\n![\"Advanced](\"https:\/\/okdiario.com\/techy\/en\/wp-content\/uploads\/2026\/08\/nvidia-huawei-deepseek-ai-chip-gap.jpg\" "\\"\\"")
Reports suggest four Huawei chips match the capability of one Nvidia processor, highlighting hardware and energy challenges in AI.<\/figcaption><\/figure>\n\n\n\nChina is building around the bottleneck<\/h2>\n\n\n\n
Huawei\u2019s answer is to make thousands of processors behave like one large machine through high-speed interconnects<\/a>. <\/p>\n\n\n\nDeepSeek, meanwhile, has leaned more heavily on Huawei hardware and is developing an inference chip of its own, according to Reuters, as it seeks more control over the infrastructure behind its models. The project is still at an early stage and faces manufacturing and memory constraints. <\/p>\n\n\n\n
This strategy can narrow the usable-computing gap even when individual Chinese accelerators remain behind Nvidia\u2019s best chips. But it also shifts the contest toward larger systems, more complex networking, and potentially heavier energy and material demands. <\/p>\n\n\n\n
\n\n\nAlso Read: Egypt suffered its first attack of the regional war after a drone hit two gas ships near the Suez Canal<\/a><\/h4>\n<\/div>\n<\/div><\/div>\n<\/div>\n\n\n\nThe likely result is that export controls<\/a> change where and how the resources are spent rather than stopping the computing race.<\/p>\n\n\n\nPerformance per watt wins<\/h2>\n\n\n\n
The meaningful environmental benchmark is not the number of chips in a press release. It is useful model output per watt, measured with the same workload and including model quality, speed, utilization, networking, cooling, and the carbon intensity of the electricity supply. Without that information, \u201cfour for one\u201d is a strategic talking point, not an ecological scorecard.<\/p>\n\n\n\n
The headline number is not four chips versus one. It is how much electricity, cooling, material<\/a>, and carbon are required for one useful unit of AI.<\/p>\n\n\n\nThe latest official statement on the Atlas 950 SuperPoD was published on Huawei\u2019s website<\/em><\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"A leaked account of a nearly four-hour investor meeting attributed to DeepSeek founder Liang Wenfeng has put an unusually blunt … <\/p>\n
That makes the comparison partly forward-looking, not a public apples-to-apples test that independent researchers can reproduce today. <\/p>\n\n\n\n Chip count does not translate directly into electricity use. Nvidia says one liquid-cooled GB300 NVL72 rack<\/a> contains 72 Blackwell Ultra graphics processing units, or GPUs, and can require up to 142 kilowatts. <\/p>\n\n\n\n If 50,000 GPUs were arranged in those full racks and operated at maximum draw, the result would be roughly 695 racks and 99 megawatts of IT equipment capacity before broader facility overhead is added.<\/p>\n\n\n\n That calculation is only an illustration, not a verified DeepSeek installation. Huawei says a full Atlas 950 SuperPoD will connect 8,192 neural processing units, or NPUs, across 160 cabinets and occupy about 10,800 square feet, but its official announcement does not state a directly comparable total power figure. <\/p>\n\n\n\n Four accelerators could use less, similar, or more electricity than one GB300 once utilization, memory, interconnects, cooling, and software efficiency enter the picture.<\/p>\n\n\n\n More chips also mean more than a larger electric bill. They can require additional circuit boards, optical links, power equipment, cooling hardware<\/a>, replacement parts, and physical space. The environmental cost may not rise in a neat four-to-one line, but it rarely stops at the processor package.<\/p>\n\n\n\n The International Energy Agency estimates that data centers consumed around 415 terawatt-hours of electricity worldwide in 2024. Its base case projects that total to reach about 945 terawatt-hours by 2030, while electricity used by accelerated servers grows around 30% each year. AI is not the only cause, but it is the most important new driver. <\/p>\n\n\n\n China and the United States are expected to account for nearly 80% of global data center electricity-demand growth through 2030. That concentration matters because massive computing campuses connect to local grids, where generation and transmission projects move much more slowly than chip product cycles. <\/p>\n\n\n\n A more efficient hardware and software path could cut data center demand by more than 15% in the IEA\u2019s high-efficiency scenario<\/a>. <\/p>\n\n\n\n DeepSeek\u2019s reputation was built on doing more with constrained hardware. Its official DeepSeek-V3 technical report describes a mixture-of-experts model with 671 billion total parameters and 37 billion activated for each token, trained with 2.788 million H800 GPU-hours. <\/p>\n\n\n\n The company estimated the direct rental cost at about $5.58 million, while explicitly excluding earlier research and experimental work. <\/p>\n\n\n\n That record helps explain why the leaked comments matter. DeepSeek\u2019s efficiency was not simply a green initiative or a clever marketing line. It was an engineering response to limited access to advanced chips, and the alleged transcript suggests Liang still sees raw computing power as the main constraint. <\/p>\n\n\n\n There is a catch. Lower energy use per training run or generated token does not guarantee lower total consumption if cheaper AI leads to many more models, users, and queries. Efficiency can slow the growth curve, but it can also make demand easier to expand.<\/p>\n\n\n\n Huawei\u2019s answer is to make thousands of processors behave like one large machine through high-speed interconnects<\/a>. <\/p>\n\n\n\n DeepSeek, meanwhile, has leaned more heavily on Huawei hardware and is developing an inference chip of its own, according to Reuters, as it seeks more control over the infrastructure behind its models. The project is still at an early stage and faces manufacturing and memory constraints. <\/p>\n\n\n\n This strategy can narrow the usable-computing gap even when individual Chinese accelerators remain behind Nvidia\u2019s best chips. But it also shifts the contest toward larger systems, more complex networking, and potentially heavier energy and material demands. <\/p>\n\n\n\n The likely result is that export controls<\/a> change where and how the resources are spent rather than stopping the computing race.<\/p>\n\n\n\n The meaningful environmental benchmark is not the number of chips in a press release. It is useful model output per watt, measured with the same workload and including model quality, speed, utilization, networking, cooling, and the carbon intensity of the electricity supply. Without that information, \u201cfour for one\u201d is a strategic talking point, not an ecological scorecard.<\/p>\n\n\n\n The headline number is not four chips versus one. It is how much electricity, cooling, material<\/a>, and carbon are required for one useful unit of AI.<\/p>\n\n\n\n The latest official statement on the Atlas 950 SuperPoD was published on Huawei\u2019s website<\/em><\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":" A leaked account of a nearly four-hour investor meeting attributed to DeepSeek founder Liang Wenfeng has put an unusually blunt … <\/p>\nThe power math is different<\/h2>\n\n\n\n
AI\u2019s electricity curve is steep<\/h2>\n\n\n\n
Also Read: The US just placed 17 states under emergency order to stop blackouts as extreme heat overloads the grid<\/a><\/h4>\n<\/div>\n<\/div><\/div>\n<\/div>\n\n\n\n
Scarcity shaped DeepSeek<\/h2>\n\n\n\n
China is building around the bottleneck<\/h2>\n\n\n\n
Also Read: Egypt suffered its first attack of the regional war after a drone hit two gas ships near the Suez Canal<\/a><\/h4>\n<\/div>\n<\/div><\/div>\n<\/div>\n\n\n\n
Performance per watt wins<\/h2>\n\n\n\n